Discussion Intel current and future Lakes & Rapids thread

[SiliconFly]

So here are the slides again: https://www.igorslab.de/en/intels-i...n-for-raptor-lake-s-refresh-and-arrow-lake-s/

It clearly says performance projection, now why on earth would they project an ES performance? Makes zero sense. The only thing missing here is clock speed, and that's it!.

AFAIK, putting a number as a goal is not how chips, nor any other engineering projects are done. The designers would clear out the low hanging fruits first then do what they can with the rest with compromises. Again, putting a number early is almost never done. Getting more performance at the last minute like with zen 1 over Excavator took a huge toll and decision like that is almost always abandoned.

Trying to make excuses to circumvent facts, which are clearly shown, is mental gymnastic and it doesn't help with the discussion at all.

Having said, even after asking a bunch of people here and on my alma mater, I still think arrow lake final performance number will be better than a meager single digit ipc uplift. It's just now possible for a company like intel to fail this hard.

Kindly check the disclaimer in igorslab article which spells it out very clearly:
"even though the finished CPU could of course perform differently"

He clearly says the projections are true, but no one knows whether it's ES or QS or retail or N3B tile or 20A tile or simulation or whether an ES even exists. It's a projection without a valid context. That's what people tried to tell.

It's way too early to speculate on ARL final performance. But general consensus is, like MTL, ARL is gonna be all about massive increase in power efficiency and probably a slight bump in performance. But like I said, it's just an educated guess, not facts.

 

[SiliconFly]

Those slides are from intel, what are you even talking about? Intel is a leaky ship, this has been known for years. Disgruntled HB1 visa holders talked and oh boy do they talk a lot.

A lot has changed in recent times. For example, MTL is barely 6 weeks away and we don't even know about it's actual power efficiency yet. The secrecy Intel maintains these days is in fact very frustrating. They're acting more like their competition now.

 

[igor_kavinski]

They're acting more like their competition now.

Pretending is a better word for what they are doing

 

[SiliconFly]

Agree. Read something interesting yesterday by Jim Keller. He said that too many things are broken inside Intel. And he also believes that Pat's strategy (5N in 4Y) is not the right way to go.

Pat might not agree with that, but all things considered, if they screw up ARL, I don't think they have a future. They can't keep doing +++++ or refresh-refresh-refresh anymore.

 

[Geddagod]

He fixed some mistakes of the past. <----- (Process related)

What?

He started the new core design (LNC & beyond) <----- (MLID's babble about royal core)

He almost certainly didn't. Maybe he oversaw some very high level decisions, or maybe he took part of the RYC vs traditional big core battles that were taking place at Intel, but he wasn't the chief architect. He was too "high level".

He made the core designs node agnostic. <----- (Starts with ARL instead of MTL)

Started with SNC

None of the ARL leaks so far are credible enough to be taken seriously. They either lack context or mere speculations or are not trustworthy enough. Same goes with the "leakers". Always trace the leaks to the source.

Both the Igors leak and the Uzi leak were credible enough. What context do you think is lacking? No context was lacking smh.

there's no way to judge the performance of LNC final as they don't even have an ES as of now.

Sure there is, just look at Intel's own final perf projections.

Kindly check the disclaimer in igorslab article which spells it out very clearly:
"even though the finished CPU could of course perform differently"

Yes, because they are projecting final perf. Not because they are simulating ES vs QS perf lol

but no one knows whether it's ES or QS

Why does that matter? At best estimations get more accurate the further in the dev process you get, but even if it was an ES1, it's highly unlikely you quadruple your final perf projections...

or N3B tile or 20A tile or simulation

ARL-S is only N3B,

It's way too early to speculate on ARL final performance.

Pretty clear it's not, since Intel themselves are lol

A lot has changed in recent times. For example, MTL is barely 6 weeks away and we don't even know about it's actual power efficiency yet. The secrecy Intel maintains these days is in fact very frustrating. They're acting more like their competition now.

Not really. Power efficiency leaks, as a whole, rarely do happen. The reasons for that are numerous, but in short MTL's leak cycle was nothing out of the ordinary.

 

[SiliconFly]

What?

Keller himself said the design group was operating without pdks. He changed that among other things. Check keller's interview.

He almost certainly didn't. Maybe he oversaw some very high level decisions, or maybe he took part of the RYC vs traditional big core battles that were taking place at Intel, but he wasn't the chief architect. He was too "high level".

I didn't say it. MLID keeps claiming constantly that RYC is Jim Keller's and it starts with LNC. Not me.

Started with SNC

I think pat said it was supposed to start with RWC. But i think it's LNC cos this is where they're dual sourcing.

Both the Igors leak and the Uzi leak were credible enough. What context do you think is lacking? No context was lacking smh.

Sure there is, just look at Intel's own final perf projections.

Yes, because they are projecting final perf. Not because they are simulating ES vs QS perf lol

Why does that matter? At best estimations get more accurate the further in the dev process you get, but even if it was an ES1, it's highly unlikely you quadruple your final perf projections...

Igor himself doesn't think it's final. Might not even be ES. No credible leaks for ARL ES yet.

ARL-S is only N3B,

Considering all the hoopla around N3B, ARL-S might also be on 20A. After all, it's node agnostic and dual sourced.

Pretty clear it's not, since Intel themselves are lol

Thats what I said too

Not really. Power efficiency leaks, as a whole, rarely do happen. The reasons for that are numerous, but in short MTL's leak cycle was nothing out of the ordinary.

Not even a single ES power usage benchmark this late in the cycle? Its definitely unlike intel.

 

[eek2121]

We don't know if that slide pertained to N3B or 20a compute tiles.



Yeah but why put an early buggy ES estimate into a slide deck? Something there doesn't add up.



How would that slide have looked if the PL2 target were 177W instead of 250W? A 7950X loses very little MT performance moving from 170W TDP/230W PPT down to ~125W PPT. But a 13900k/ks/14900k loses more . If Arrow Lake doesn't scale well with increased PL2 values then it may still be relatively competent @177W PL2.

So here are the slides again: https://www.igorslab.de/en/intels-i...n-for-raptor-lake-s-refresh-and-arrow-lake-s/

It clearly says performance projection, now why on earth would they project an ES performance? Makes zero sense. The only thing missing here is clock speed, and that's it!.

AFAIK, putting a number as a goal is not how chips, nor any other engineering projects are done. The designers would clear out the low hanging fruits first then do what they can with the rest with compromises. Again, putting a number early is almost never done. Getting more performance at the last minute like with zen 1 over Excavator took a huge toll and decision like that is almost always abandoned.

Trying to make excuses to circumvent facts, which are clearly shown, is mental gymnastic and it doesn't help with the discussion at all.

Having said, even after asking a bunch of people here and on my alma mater, I still think arrow lake final performance number will be better than a meager single digit ipc uplift. It's just now possible for a company like intel to fail this hard.

I don’t claim to know the origins of the slides, but the context could be related to power consumption and scaling vs. perf/watt (or possibly heat dissipation).

That being said, I expect Arrow Lake won’t be that much faster than Raptor Lake refresh (at least for 1T), but perf/watt should see significant improvements.

The cumulative effect of minor increases for 3 years adds up, however, and I don’t think Intel will be behind the competition by any stretch.

 

[Goop_reformed]

Kindly check the disclaimer in igorslab article which spells it out very clearly:
"even though the finished CPU could of course perform differently"

He clearly says the projections are true, but no one knows whether it's ES or QS or retail or N3B tile or 20A tile or simulation or whether an ES even exists. It's a projection without a valid context. That's what people tried to tell.

It's way too early to speculate on ARL final performance. But general consensus is, like MTL, ARL is gonna be all about massive increase in power efficiency and probably a slight bump in performance. But like I said, it's just an educated guess, not facts.

Mind you those slides are projected performance, for the final product, not some kind of ES. Why on earth would anyone project an ES performance number <- this is where your argument breaks down.

A lot has changed in recent times. For example, MTL is barely 6 weeks away and we don't even know about it's actual power efficiency yet. The secrecy Intel maintains these days is in fact very frustrating. They're acting more like their competition now.

A lot has changed for the worse aka delays + unimpressive performance numbers. Power efficiency obviously goes up because raptor lake drinks power for breakfast.

I don’t claim to know the origins of the slides, but the context could be related to power consumption and scaling vs. perf/watt (or possibly heat dissipation).

That being said, I expect Arrow Lake won’t be that much faster than Raptor Lake refresh (at least for 1T), but perf/watt should see significant improvements.

The cumulative effect of minor increases for 3 years adds up, however, and I don’t think Intel will be behind the competition by any stretch.

The slides are legit. As mention before the only missing info is the clock. Power consumption is closely related to tdp when it comes to intel. It still might be wrong because of the denser process though.

 

[SiliconFly]

Mind you those slides are projected performance, for the final product...

What makes you think that it's final when Igor himself says otherwise? You got better insight than Igor?

 

[Goop_reformed]

What makes you think that it's final when Igor himself says otherwise? You got better insight than Igor?

Because I read

"However, I have been given internal performance projections that are supposed to forecast the performance of a current Core i9-13900K with the appropriate models of Raptor Lake Refresh and Arrow Lake. Since this is not public PR material, these projections can certainly be considered realistic, even though the finished CPU could of course perform differently. But at least it is more than just a rough benchmark."

 

[SiliconFly]

"However, I have been given internal performance projections that are supposed to forecast the performance of a current Core i9-13900K with the appropriate models of Raptor Lake Refresh and Arrow Lake. Since this is not public PR material, these projections can certainly be considered realistic, even though the finished CPU could of course perform differently. But at least it is more than just a rough benchmark."

Igor thinks different.

Btw, the slide you keep referring to seems to have been cooked up by some Intel marketing guy who has no inkling of engineering. ARL-S is a 8+16+2 product, not a 8+16+1 as mentioned in that slide.

 

[coercitiv]

Btw, the slide you keep referring to seems to have been cooked up by some Intel marketing guy who has no inkling of engineering. ARL-S is a 8+16+2 product, not a 8+16+1 as mentioned in that slide.

The last number is not designating the new E cores, but rather the GPU config. Raptor Lake is also marked as 8+16+1.

 

[Goop_reformed]

Igor thinks different.

Btw, the slide you keep referring to seems to have been cooked up by some Intel marketing guy who has no inkling of engineering. ARL-S is a 8+16+2 product, not a 8+16+1 as mentioned in that slide.

You are deliberately avoiding to address the issue here.

Igor thinks this is final performance since it's not pr slides. "Of course it could perform differently" is an asterisk, a disclaimer if you will since final product has not been released.

And how is this made by marketing guy? and for who exactly? Stop skirting the main point Igor made. Being positive is one thing but refusing to take the those infos at face values is absolutely disingenuous.

 

[SiliconFly]

You are deliberately avoiding to address the issue here.

Igor thinks this is final performance since it's not pr slides. "Of course it could perform differently" is an asterisk, a disclaimer if you will since final product has not been released.

And how is this made by marketing guy? and for who exactly? Stop skirting the main point Igor made. Being positive is one thing but refusing to take the those infos at face values is absolutely disingenuous.

This is a free world. Feel free to believe in what you choose to.

I'm not going to base any of my discussions on one questionable slide.

 

[Goop_reformed]

This is a free world. Feel free to believe in what you choose to.

I'm not going to base any of my discussions on one questionable slide.

The slides are not questionable, at all. You'd think Igor want to confirm the content with his intel contact before publishing anything. Moreover, this is not about what I believe in. This about these slides and your weird obsession with twisting it in accordance to your personal bias.

And there are MULTIPLE slides, not just from Igor and uzzi in regards to Arrow Lake. This has been know within the industry for more than 6 months by now.

And I, have stated for multiple times, that final performance number should definitely higher. It's just you keep repeating the same fringe conclusion even though all the evidence presented dictate otherwise. This is neither helpful nor technical.

Again, I'm happy to discuss new slides from lenovo, when you're in possession of them that is.

 

[H433x0n]

Again, I'm happy to discuss new slides from lenovo, when you're in possession of them that is.

Do slides give any information on fmax?

 

[ondma]

Because I read

"However, I have been given internal performance projections that are supposed to forecast the performance of a current Core i9-13900K with the appropriate models of Raptor Lake Refresh and Arrow Lake. Since this is not public PR material, these projections can certainly be considered realistic, even though the finished CPU could of course perform differently. But at least it is more than just a rough benchmark."

Did you even read that paragraph? Unfortunately, I think ARL will be disappointing, especially if they cant even get hyperthreading or rental units working, but nowhere in that paragraph does it state or imply that those figures cannot change.

 

[Goop_reformed]

Do slides give any information on fmax?

It talks about tdp so in some sense, yes.

Did you even read that paragraph? Unfortunately, I think ARL will be disappointing, especially if they cant even get hyperthreading or rental units working, but nowhere in that paragraph does it state or imply that those figures cannot change.

Do you? And by the way, final product can perform lower than projection, have you considered that?

What important is to take the slides at face values. "even though the finished CPU could of course perform differently" = author anecdote which is common sense. All leak videos have this.

 

[S'renne]

It talks about tdp so in some sense, yes.



Do you? And by the way, final product can perform lower than projection, have you considered that?

What important is to take the slides at face values. "even though the finished CPU could of course perform differently" = author anecdote which is common sense. All leak videos have this.

Pretty sure if its worse they wouldn't lower the P2 to 177W and would instead crank it higher

 

[Goop_reformed]

Pretty sure if its worse they wouldn't lower the P2 to 177W and would instead crank it higher

"finished CPU could of course perform differently" I'm only repeating this because people keep want to emphasize this bit. Final performance could be higher or lower. See how useless this is?

 

[msj10]

this image was posted by xinoassasin on twitter.
I think the igor slides are from these documents that intel releases to OEMs. PPG is power performance guide, but I don't know what presample is.

 

[igor_kavinski]

View attachment 88106

this image was posted by xinoassasin on twitter.

Wish some Robinhood hacker would penetrate into an OEM network and leak these

 

[igor_kavinski]

Pretty sure if its worse they wouldn't lower the P2 to 177W and would instead crank it higher

I strongly think they CAN'T

As in, any higher leads to too much leakage current and heat output with maybe 1% or 2% higher performance. Not worth it.

 

[Geddagod]

View attachment 88106

this image was posted by xinoassasin on twitter.
I think the igor slides are from these documents that intel releases to OEMs. PPG is power performance guide, but I don't know what presample is.

im guessing that's where most leakers prob get their info tbh

 

[SiliconFly]

It talks about tdp so in some sense, yes.

And you believe you can deduce ARL's Fmax based on it's TDP? Interesting!!!

And by the way, final product can perform lower than projection...

Do you even believe in your own words? Cos' its exciting.

It's very interesting that you want to declare ARL a lost cause based on one quirky slide. But things work a bit differently here. Unless we have multiple credible sources and information properly vetted, no one here is gonna take it seriously.